虚拟图像特征融合误差控制算法仿真  

Simulation of Error Control Algorithm for Virtual Image Feature Fusion

作　　者：庄佳 石林[1] ZHUANG Jia;SHI Lin(Changzhou University,Changzhou Jiangsu 213164,China)

机构地区：[1]常州大学,江苏常州213164

出　　处：《计算机仿真》2023年第4期219-222,230,共5页Computer Simulation

摘　　要：为使VR图像呈现更大的视场角,需利用镜头反畸变方法优化图像视觉效果,此过程易发生特征畸变。为及时矫正特征畸变,提出VR图像特征融合误差半监督校正算法。提取图像特征,获取误差定位区域。在定位区域中预测特征融合误差。误差数据在支持向量机的辅助下,通过半监督式学习算法对分类器完成训练,分类所有误差特征,最终计算出每种误差的校正系数,实现VR图像特征融合的误差校正。实验结果表明,所提算法的校正准确率始终稳定在90%~100%,在100组迭代实验中所提方法的误差预测精度可达90%以上,校正所需时间更短,均值为0.20s。In order to make VR images present a larger field of view,it is necessary to use the lens anti-distortion method to optimize the image visual effect,which is prone to feature distortion.In order to correct the feature distortion in time,this paper presented a semi-supervised correction algorithm for VR image feature fusion error.Firstly,we extracted the image features and obtained the error location area.Then we predicted feature fusion error in this region.With the help of support vector machine,the error data were trained by semi-supervised learning algorithm to classify all error features.Finally,we calculated the correction coefficient of each error,and thus,realized the error correction for VR image feature fusion.Experimental results show that the correction accuracy of the proposed algorithm is always kept betiween 90%~100%.In one hundred iterative experiments,the error prediction accuracy can reach above 90%,and the correction time is shorter,with an average of 0.20s.

关 键 词：图像 误差预测 特征提取 半监督 校正算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]